Do Readmissions Cost More than Non-Readmissions?

One of the challenges in working in healthcare data analytics is the requirement to keep up-to-date on many diverse topics. The eHealth Initiative provides a multitude of such opportunities, and we enjoyed attending the recent National Forum on Data and Analytics at which we heard a wide variety of presentations on many topics related to the management and analysis of healthcare data.

In one particularly interesting session the speaker was describing research that she had performed into causes of readmissions. Although CMS has implemented penalties for "excess readmissions" in certain clinical areas, the conventional wisdom appears to be that the cost of the current penalties is lower than the loss of Medicare revenue that would occur from reductions of the readmissions themselves. The speaker, however, suggested that in many cases the cost to the hospital (not the cost to Medicare) of a readmission exceeded the Medicare payment for that admission, and therefore it was in the hospital's financial interest to attempt to eliminate these readmissions wherever possible. Since we always enjoy opportunities to perform interesting new data analyses, we put this project on our list of interesting analytics projects that no one is paying us to perform.

When a few free moments appeared on our schedule, we dug into this question. Our first decision was to avoid the whole issue of cost versus revenue, since the conclusion depends entirely on the definition of "cost". A much easier analysis, which would still be relevant to the issue, would be to compare the costs of "index" admissions (defined as those admissions not preceded by another admission within the previous 30 days) with the costs of "readmissions", which were preceded by another admission within 30 days. And to further simplify the problem, we utilized charges rather than attempting to estimate costs. This was done for two reasons – first, because charge information was easily available from our data source, and also because computing any type of cost figure from this data will be time-consuming and would not provide significant additional value.

At this point we should note that our objective is not to present a scientific statistical analysis of this issue, but rather to demonstrate some anecdotal results that we found when analyzing this interesting issue. Perhaps a more research-oriented organization will be intrigued by this analysis and apply more robust statistical methods that might generate different results.

Analytical Process

Fortunately, the Singletrack/DataGen team has available the Medicare 100% standard analytic files for 2011. These files contain 100% of all Medicare claims for that period, except for Part B claims which fortunately were not relevant to this project. So we wrote some SQL code that would tag each admission that was preceded by another admission for the same beneficiary within the preceding 30 days. We separated this data by DRG and provider identifier and included the charges and Medicare payment amounts from claims. This resulted in about 12 million records of inpatient claims, which we decided was an adequate sample for our analysis. We dumped those claims into the mighty Excel PowerPivot analytics engine and started digging into the data.

We decided that the most significant metric to measure here was the ratio of average charges for readmission to the average charge for index admissions within the same DRG. The accuracy this metric would be affected, of course, by differences in charges among hospitals, so an important step would be to stratify the data by hospital to get consistent conclusions. We focused on the DRGs included in the CMI readmission penalty (AMI, heart failure, pneumonia), but also looked at some other high-volume DRGs.

Results of analysis

The table below presents some initial results. For each DRG it shows the ratio of the average readmission charge to the average index admission charge. The ratios are color-coded to facilitate the analysis. Also included is the count of admissions in each DRG, which will indicate the relative prevalence of the DRGs.

As can be noted, results vary significantly across groups of DRGs. For AMI patients, the charges for readmissions slightly exceeded those of index admissions for DRG 280, but were lower for DRGs 281 and 282. For heart failure readmissions have slightly higher charges, and for pneumonia they have significantly higher charges.

The relationships shown in the table above may be reasonable, but they suffer from the differences in hospital charge rates described above. Therefore, we prepared the graphs below indicating the charge relationships for 50 high-volume hospitals for the major DRGs shown above. In these graphs, each vertical bar represents the charge relationship for an individual hospital. Since charges within the hospital are consistent, these relationships should be reliable. The red line shows the number of admissions that the respective hospital had in that DRG.

Also notable from this data is the variation among hospitals of the ratio. In some hospitals readmissions appear to be significantly more costly than index admissions, while another hospitals the converse is true.

Finally, there are significant difference in charges per admission within hospitals. The graph below shows the variation in charges within one hospital across all admissions, with the index admissions shown in blue bars and readmissions shown in red. As can be seen, there are significant variation in charges between admissions, for both readmissions and index admissions. Therefore, this relationship in any individual hospital may differ significantly from those presented here.

Conclusions

Astute readers will realize that we haven’t answered real question here, which is whether the net financial effects to hospitals of readmission reductions (including the loss of revenue and the Medicare penalties) will be positive or negative. That's because the Medicare penalties are applied to all Medicare admissions, and that analysis was beyond this project's scope. But we have created a partial answer, in that for certain DRGs hospital charges for readmissions are significantly higher than for index admissions; therefore it may be plausible that these the cost of these readmissions exceed their respective Medicare revenue. (That again takes us into the definition of "cost", which is also beyond our scope here.) If we have found no difference in average charges between readmissions and index admissions, the entire concept would have been implausible.

Areas for future analysis

Although we've reached the end of our analytical journey here, this area maybe fertile ground for statisticians looking for interesting opportunities to apply their skills. We haven't tested for significance of difference between readmission and index admission charges, for example, nor have we dealt with differences in sample size between those two populations. Perhaps additional, more robust analysis would lead to different conclusions, or will ratify our results.

One further note

Our friends at DataGen suggested that one shortcut to estimating costs from charges is to use the overall hospital ratio of cost to charges. This isn't as accurate as using RCCs by cost center, but did help to equalize the results across hospitals. So we dumped the hospital RCCs into the analysis and revised a few of the graphs. The graph below shows the readmit-to-index admission ratios when calculated with charges (blue lines) and with estimated cost (red lines) for the selected hospitals. While there were some differences among the ratios, the overall conclusions for each DRG didn't change.